Why We Vote
Anyone who believes recent polls have shown them which candidate is ahead in Wisconsin's recall race has been grossly misled by media reports attributing accuracy to poll results that simply does not exist.
I say that as someone who was trained at Northwestern University to set up and interpret The Milwaukee Journal's professional polling operations.
We learned from the experts, including Philip Meyer, the author of Precision Journalism, the book that popularized newspaper polling around the country, and Bill Schneider, who at CNN did the same for television networks.
The people who know the most about polling are most appalled by sloppy reporting and missing caveats that produce misleading headlines accepted as fact.
Some journalists today are knowledgeable enough about the pitfalls of polling to include the margin of error and known partisan leanings of a particular pollster.
But those qualifications instantly disappear in the minds of readers or viewers. And serious flaws from distorted samples are rarely reported at all.
Too Close to Call
Horse races are most exciting when you can see one of the horses coming up fast or starting to pull away. When the horses are all clustered together, there's nothing much to see.
The most interesting fact about the recall race between Barrett and Walker confirmed by all the polls is that it's so incredibly close.
It's probably much closer than you think. And none of the polls really tells you which candidate is ahead. But reporters need a clear narrative.
So when a Marquette University Law School poll in April showed Barrett with a small lead over Walker, 48% to 47%, and the next Marquette poll two weeks later showed Walker leading Barrett, 50% to 44%, the story was that Walker had overcome a tiny deficit and opened up a small lead.
Never mind that both polls were exactly the same. The difference in both was within the margin of error. So neither really told you which candidate was ahead.
The Marquette poll has been given outsize importance in this recall despite legitimate reasons for caution.
First of all, it is a brand-new poll created by a conservative law school. That means it has no history by which anyone can judge its credibility or accuracy.
Perhaps to reassure the public that its conservative leanings wouldn't influence the results, the school hired Charles Franklin, a political science professor and polling expert from the University of Wisconsin-Madison, to create and analyze its polls.
Students say Franklin identifies himself in his classes as a Republican, but that is no reason to doubt his professional standards.
I stopped overseeing Journal polling when I started writing a brazenly liberal opinion column, but that had more to do with public perception and nervous editors. I had no doubt I could separate the two jobs and do them both honestly.
The public frequently suspects partisan pollsters of weighting polls in the direction they want them to go, in order to influence public opinion instead of reporting it.
But any party professional who doesn't want to know accurately how a campaign is really going is an idiot. So most do as much accurate polling internally as possible, and then only release the results that make them look good.
Still, insiders looking at the Marquette poll sample have raised some pointed concerns.
First of all, the poll under-sampled Barrett voters, looking at the 2010 election between the same two candidates. In that Tea Party-driven election, Walker beat Barrett 52% to 46%. In Marquette's poll, fewer than 41% say they voted for Barrett.
That pretty much accounts for the entire lead the poll claims for Walker. The poll also over-sampled Republicans based on age (45% over 60) and geography.
Based on statewide population and voting patterns, an accurate sample would have included only 12% from strongly Republican Waukesha, Washington and Ozaukee counties and more than 17% from the city of Milwaukee and Dane County, which lean Democratic.
Instead, Marquette polled 30% from Waukesha, Washington and Ozaukee counties and 29% from the city of Milwaukee and Dane County.
The hardest part for any professional poll, especially a brand-new poll, to determine is likely voters. That is almost impossibly difficult in the current recall because there has never been a comparable election in Wisconsin.
So believe those disclaimers describing the poll differences between the candidates as "statistically insignificant."
That's why Democrats are eagerly targeting 400,000 Wisconsinites who voted in President Obama's landslide 2008 victory, but didn't vote in the 2010 off-year election. Internal polling shows Barrett leads 57% to 37% among those voters.
Polls won't decide the recall. Determined voters will do that on June 5.